Research on equivalent modeling and model testing verification methods for material mechanics parameters of TXV structure

Abstract

Aiming at the problems of multi-scale mesh division and low computational efficiency encountered during TXV (through X via) structural simulation, an equivalent modeling method for material mechanics parameters of TXV structure is proposed. By conducting mechanics simulations on the minimum elements of three TXV structures, namely through silicon via (TSV), through mold via (TMV) and through glass via (TGV), the anisotropic material mechanics parameters of TXV structures corresponding to different substrate materials, via diameters, via depths and via pitches are obtained. Based on simulation data and BP (back propagation) neural network algorithm, the material mechanics parameter equivalence models of TXV structures are established. Based on the four-point bending method, the theoretical calculation and simulation analysis of the actual TSV structure are carried out, and the accuracy of the equivalent model is tested and verified by the force and strain measurement systems. The simulation and test results show that the prediction accuracy of equivalent models based on BP algorithm is high, and the prediction errors for training samples of TSV, TMV and TGV structures are less than 0.467 %, 3.571 % and 1.303 %, respectively, and the prediction errors for testing samples are less than 0.424 %, 3.130 % and 1.444 %, respectively. After equivalence and simplification, the mesh division and computational efficiency of the simulation model are improved significantly, the number of mesh elements is reduced by 95.79 %, and the calculation time is shortened by 94.17 %. The simulation accuracy based on the equivalent model is high, the strain simulation error between the simplified model and the complete model is 0.24 %, and the error between the strain simulation result of the simplified model and the actual measurement result is 2.81 %.